Comparative gene expression profiling identifies common molecular signatures of NF-κB activation in canine and human diffuse large B cell lymphoma (DLBCL)

Abstract

We present the first comparison of global transcriptional changes in canine and human diffuse large B-cell lymphoma (DLBCL), with particular reference to the nuclear factor-kappa B (NF-κB) pathway. Microarray data generated from canine DLBCL and normal lymph nodes were used for differential expression, co-expression and pathway analyses, and compared with analysis of microarray data from human healthy and DLBCL lymph nodes. The comparisons at gene level were performed by mapping the probesets in canine microarrays to orthologous genes in humans and vice versa. A considerable number of differentially expressed genes between canine lymphoma and healthy lymph node samples were also found differentially expressed between human DLBCL and healthy lymph node samples. Principal component analysis using a literature-derived NF-κB target gene set mapped to orthologous canine array probesets and human array probesets clearly separated the healthy and cancer samples in both datasets. The analysis demonstrated that for both human and canine DLBCL there is activation of the NF-κB/p65 canonical pathway, indicating that canine lymphoma could be used as a model to study NF-κB-targeted therapeutics for human lymphoma. To validate this, tissue arrays were generated for canine and human NHL and immunohistochemistry was employed to assess NF-κB activation status. In addition, human and canine B-cell lymphoma lines were assessed for NF-κB activity and the effects of NF-κB inhibition.

Figure 1. PCA of global gene expression profiles of canine and human datasets.

Three dimensional plots of the PCA of canine and human datasets using the first three principal components of global gene expressions. The PCAs show two distinct clusters in both the datasets: the healthy cluster and DLBCL cluster. Blue spheres denote the healthy samples and red spheres denote DLBCL samples. An ellipse was drawn around each of the clusters to mark the limit of the distance of 3 standard deviations from the centre. (A) In the canine dataset (GSE30881), the healthy cluster has all the 10 healthy samples and the cancer cluster has all the 23 canine DLBCL samples. The amounts of variance captured by the first three principal components are 22.7%, 6.97% and 6.75% respectively. (B) In the human dataset (GSE12195), the healthy cluster has all the 10 healthy samples and the cancer cluster has all the 45 human DLBCL samples. The amounts of variance captured by the first three principal components are 17.3%, 14.7% and 6.6% respectively.

Figure 2. PCA of canine and human datasets using exclusively the expression levels of the NF-κB target genes (probesets).

Three dimensional plots of the PCA of canine and human datasets using the first three principal components of the expression levels of the NF-κB target genes (probesets) show clear separation of DLBCL samples from the healthy samples in both the datasets. Blue spheres denote the healthy samples and red spheres denote the DLBCL samples. The ellipsoids drawn around the clusters mark the limit of the 2 standard deviations from the centre in 3-dimensional space. (A) In the canine dataset (GSE30881), the amounts of variance captured by the first three principal components are 23.6%, 14% and 8.52% respectively. (B) In the human dataset (GSE12195), the amounts of variance captured by the first three principal components are 31.6%, 7.87% and 6.54% respectively.

Figure 3. Hierarchical clustering of canine and human datasets using exclusively the expression levels of the NF-κB target genes (probesets).

Hierarchical clustering of the canine (A) and human (B) datasets using exclusively the expression levels of the NF-κB target gene set. The samples are arranged in the columns (blue squares denote healthy and red squares denote DLBCL) and the probesets are in the rows. The dendrograms are drawn using Euclidean distances with average linkage method. (A) In the canine dataset (GSE30881), the 199 NF-κB target probesets separate the dataset into three top-level clusters. While the first and the third clusters have exclusively of DLBCL samples, the second cluster has all the healthy samples with two DLBCL samples. (B) In the human dataset (GSE12195), the 259 NF-κB target probesets separate the dataset into two top-level clusters. The first cluster has 12 samples that include all the healthy samples and two DLBCL samples, while the second cluster is solely of 43 DLBCL samples. The numbers above each column refer to sample identification numbers.

Figure 4. Comparison of differentially expressed probesets in canine and human DLBCLs.

Venn diagrams comparing the NF-κB target genes in the differentially expressed probesets (between DLBCL and healthy) of canine DLBCL and human DLBCL and comparison of the number of the differentially expressed probesets of canine DLBCL and human DLBCL. The differentially expressed probesets in the datasets were identified by one-way ANOVA (DLBCL Vs. healthy) of the expression values; selecting probesets with log₂ fold change over 2 with FDR adjusted p-value less than 0.05. (A) 25 NF-κB target probsets (17 NF-κB target genes out of the 120 genes) present in the differentially expressed probesets of canine DLBCL. (B) 101 NF-κB target probsets (54 NF-κB target genes out of the 120 genes) present in the differentially expressed probesets of human DLBCL. (C) Comparison of canine array probesets converted to orthologous human array probesets. (D) Comparison of human array probesets converted to orthologous canine array probesets.

Figure 5. Comparison of enrichment of KEGG pathways in the co-expressed clusters of canine DLBCL and human DLBCL.

The highly connected gene clusters identified in the co-expression networks of canine DLBCL and human DLBCL were analysed for enrichment of KEGG pathways using DAVID functional annotation tool. The results from the analysis of each cluster are compiled and the p-values of the enrichment score computed by Fisher's exact test are represented graphically as coloured icons.

Figure 6. Comparison of enrichment of signalling pathways in the co-expression clusters of canine DLBCL and human DLBCL using IPA.

The highly connected gene clusters identified in the co-expression networks of canine DLBCL and human DLBCL were analysed for enrichment of lymphoma related signalling pathways using IPA tool. The result of the comparison analysis is represented graphically. The numbers on the bars denote the co-expression cluster numbers while the length of the bars show their significance, negative log of the p-value for the pathway computed by Fisher's exact test, in the relevant pathway.

Figure 7. In vivo and in vitro Validation of NF-κB Status in Canine and Human DLBCL.

NF-κB expression and activation in human T and B cell lymphoma cell lines and canine B cell lymphoma cell line. Whole cell lysates from Jurkat, JM1, Pfeiffer, RL and 3132 cells were eletrophoresed and immunoblotted for the detection of expression of signaling components of the NF-κB pathway (A). Nuclear protein from human T (Jurkat) and B (RL, JM1, Pfeiffer) cell lymphoma lines as well as canine B (3132) cell lymphoma line, were extracted. EMSAs were performed on these samples using non-radioactive DIG-labeled NF-κB consensus oligonucleotide probes in binding reactions. Specificity was tested by competition using unlabeled (cold) specific and non-specific probes and “supershift” using specific antibody for NF-κB p65/RelA and a non-specific antibody (p53). Samples were then subjected to electrophoresis in DNA retardation gels, before transfer onto nylon membranes and chemiluminescence detection of the DIG-labels. EMSAs were performed using nuclear extracts of canine (B) and human (C) lymphoma cell lines. Bands are indicated by arrows and annotated. ‘+’ and ‘−‘ refer to components being added or omitted respectively, to standard gel shift binding reactions.

Figure 8. Illustrative photomicrographs showing different nuclear staining intensities for p65 and p52.

p65 staining in human DLBCL; weak (a), strong (b). p65 staining in canine DLBCL weak (c), strong (d). p52 staining in human DLBCL; weak (e), strong (f). p52 staining in canine DLBCL; weak (g), strong (h).

Figure 9. Sensitivity of 3132 cells to NF-κB inhibitor, IKK inhibitor VII, doxorubicin and vincristine.

Cell viability of 3132 cells were measured post-treatment with NF-κB inhibitor (A), doxorubicin (B) or vincristine (C) in the presence/absence of IKK inhibitor VII or with IKK inhibitor VII only, using Promega CellTiter 96® AQ_ueous One Solution Cell Proliferation MTT assays. Data from doxorubicin and vincristine single treatments are depicted in B and C respectively as small closed diamond points on short dashed lines while IKK inhibitor VII (B and C) and NF-κB inhibitor (A) single treatments are represented as large closed square points on long dashed lines. Combination treatments of doxorubicin (B) or vincristine (C) with IKK inhibitor VII are presented as open triangular points on solid lines.

Figure 10. The effect of doxorubicin and IKK inhibitor VII on the activation of NF-κB in human B cell lymphoma cell lines and canine B cell lymphoma cell lines.

Nuclear protein from human B (RL, JM1, Pfeiffer) cell lymphoma lines as well as canine B (3132) cell lymphoma line treated with doxorubicin (A) and/or IKK inhibitor VII (B), were extracted. EMSAs were performed on these samples using non-radioactive DIG-labelled NF-κB consensus oligonucleotide probes in binding reactions. Samples were then subjected to electrophoresis in DNA retardation gels, before transfer onto nylon membranes and chemiluminescence detection of the DIG-labels. Bands are indicated by arrows and annotated. ‘+’ and ‘−’ refer to components being added or omitted respectively, to standard gel shift binding reactions. Expression of classical pathway NF-κB subunits in 3132 nuclear extracts when 3132 cells are treated with doxorubicin and/or IKK inhibitor VII at IC₅₀ doses was detected by western blotting of whole cell lysates (B).